摘要
基于高光谱成像技术结合模式识别,建立了苹果表面缺陷识别模型。首先,利用高光谱图像采集系统采集完好无损和表面有缺陷苹果的高光谱图像,提取感兴趣区域的平均光谱反射率;然后,比较标准正态变换(SNV)和多元散射校正(MSC) 2种光谱预处理方法对建模效果的影响,得出MSC为建模最优预处理方法。最后,采用主成分分析法选择累计贡献率超过99%的前5个主成分作为样本集特征光谱数据,分别建立了基于K最近邻(KNN)模式识别和偏最小二乘判别分析(PLS-DA)识别模型。结果表明:光谱经MSC预处理后,基于PLS-DA建立的识别模型对校正集和检验集识别率均达到100%,表明基于高光谱成像技术结合模式识别可实现苹果表面缺陷的无损检测。
The recognition model of defect on apples is established,based on hyperspectral imaging technology combined with pattern recognition.Firstly,the hyperspectral imaging system is used to collect the hyperspectral image of apples with no defect and surface defect,and the average spectral reflectance of the region of interest is acquired.Then,by comparing the effects of two spectra pretreatment methods of standard normal variate(SNV)and multi-scatter calibration(MSC),it is concluded that MSC is the best pretreatment method for modeling.Finally,the first 5 principal components with cumulative contribution rate of 99%are selected as the characteristic spectral data in the sample set by principal component analysis(PCA),and K nearest neighbor(KNN)and partial least-square discriminant analysis(PLS-DA)are applied to build recognition model,respectively.The results show that the noise of the hyperspectral image of apples can be effectively removed by MSC processing,and the correct identification rates predicted by PLS-DA recognition model for apples in calibration set and prediction set both reach to 100%.This study indicates that hyperspectral imaging technology combined with pattern recognition is effective for identifying defect on apples.
作者
孟庆龙
张艳
尚静
MENG Qinglong;ZHANG Yan;SHANG Jing(Food and Pharmaceutical Engineering Institute,Guiyang University,Guiyang 550005;The Research Center of Nondestructive Testing for Agricultural Products,Guiyang University,Guiyang 550005)
出处
《食品工业》
CAS
北大核心
2019年第3期131-134,共4页
The Food Industry
基金
国家自然科学基金项目(61505036)
贵州省普通高等学校工程研究中心(黔教合KY字[2016]017)
贵州省科技厅基金项目(黔科合基础[2019]1010)
贵阳市科技局贵阳学院专项资金(GYU-KYZ[2018]01-15)
关键词
高光谱成像
模式识别
苹果缺陷
无损检测
hyperspectral imaging
pattern recognition
apple defect
nondestructive recognition
